Kan Extensions for Program Optimisation
https://www.cs.ox.ac.uk/people/ralf.hinze/publications/MPC12.pdf
Many program optimisations involve transforming a program in direct style to an equivalent program in continuation-passing style. This paper investigates the theoretical underpinnings of this transformation in the categorical setting of monads. We argue that so-called absolute Kan Extensions underlie this program optimisation. It is known that every Kan extension gives rise to a monad, the codensity monad, and furthermore that every monad is isomorphic to a codensity monad. The end formula for Kan extensions then induces an implementation of the monad, which can be seen as the categorical counterpart of continuation- passing style. We show that several optimisations are instances of this scheme: Church representations and implementation of backtracking using success and failure continuations, among others. Furthermore, we develop the calculational properties of Kan extensions, powers and ends. In particular, we propose a two-dimensional notation based on string diagrams that aims to support effective reasoning with Kan extensions.
Many program optimisations involve transforming a program in direct style to an equivalent program in continuation-passing style. This paper investigates the theoretical underpinnings of this transformation in the categorical setting of monads. We argue that so-called absolute Kan Extensions underlie this program optimisation. It is known that every Kan extension gives rise to a monad, the codensity monad, and furthermore that every monad is isomorphic to a codensity monad. The end formula for Kan extensions then induces an implementation of the monad, which can be seen as the categorical counterpart of continuation- passing style. We show that several optimisations are instances of this scheme: Church representations and implementation of backtracking using success and failure continuations, among others. Furthermore, we develop the calculational properties of Kan extensions, powers and ends. In particular, we propose a two-dimensional notation based on string diagrams that aims to support effective reasoning with Kan extensions.